KVM: Non-atomic interrupt injection

Change the interrupt injection code to work from preemptible, interrupts
enabled context.  This works by adding a ->cancel_injection() operation
that undoes an injection in case we were not able to actually enter the guest
(this condition could never happen with atomic injection).

Signed-off-by: Avi Kivity <avi@redhat.com>

4 files changed, 40 insertions, 20 deletions

diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index b43686a4487..80224bf5d4f 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -552,6 +552,7 @@ struct kvm_x86_ops {
 	void (*queue_exception)(struct kvm_vcpu *vcpu, unsigned nr,
 				bool has_error_code, u32 error_code,
 				bool reinject);
+	void (*cancel_injection)(struct kvm_vcpu *vcpu);
 	int (*interrupt_allowed)(struct kvm_vcpu *vcpu);
 	int (*nmi_allowed)(struct kvm_vcpu *vcpu);
 	bool (*get_nmi_mask)(struct kvm_vcpu *vcpu);
diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
index 1d2ea65d353..1a85fc507cf 100644
--- a/arch/x86/kvm/svm.c
+++ b/arch/x86/kvm/svm.c
@@ -3261,6 +3261,17 @@ static void svm_complete_interrupts(struct vcpu_svm *svm)
 	}
 }
 
+static void svm_cancel_injection(struct kvm_vcpu *vcpu)
+{
+	struct vcpu_svm *svm = to_svm(vcpu);
+	struct vmcb_control_area *control = &svm->vmcb->control;
+
+	control->exit_int_info = control->event_inj;
+	control->exit_int_info_err = control->event_inj_err;
+	control->event_inj = 0;
+	svm_complete_interrupts(svm);
+}
+
 #ifdef CONFIG_X86_64
 #define R "r"
 #else
@@ -3631,6 +3642,7 @@ static struct kvm_x86_ops svm_x86_ops = {
 	.set_irq = svm_set_irq,
 	.set_nmi = svm_inject_nmi,
 	.queue_exception = svm_queue_exception,
+	.cancel_injection = svm_cancel_injection,
 	.interrupt_allowed = svm_interrupt_allowed,
 	.nmi_allowed = svm_nmi_allowed,
 	.get_nmi_mask = svm_get_nmi_mask,
diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index 3237f6cc930..70af3db372d 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -3895,6 +3895,16 @@ static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
 				  IDT_VECTORING_ERROR_CODE);
 }
 
+static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
+{
+	__vmx_complete_interrupts(to_vmx(vcpu),
+				  vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
+				  VM_ENTRY_INSTRUCTION_LEN,
+				  VM_ENTRY_EXCEPTION_ERROR_CODE);
+
+	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
+}
+
 /*
  * Failure to inject an interrupt should give us the information
  * in IDT_VECTORING_INFO_FIELD.  However, if the failure occurs
@@ -4348,6 +4358,7 @@ static struct kvm_x86_ops vmx_x86_ops = {
 	.set_irq = vmx_inject_irq,
 	.set_nmi = vmx_inject_nmi,
 	.queue_exception = vmx_queue_exception,
+	.cancel_injection = vmx_cancel_injection,
 	.interrupt_allowed = vmx_interrupt_allowed,
 	.nmi_allowed = vmx_nmi_allowed,
 	.get_nmi_mask = vmx_get_nmi_mask,
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index e7198036db6..a465bd29f38 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -5005,7 +5005,6 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
 	int r;
 	bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
 		vcpu->run->request_interrupt_window;
-	bool req_event;
 
 	if (vcpu->requests) {
 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
@@ -5041,6 +5040,21 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
 	if (unlikely(r))
 		goto out;
 
+	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
+		inject_pending_event(vcpu);
+
+		/* enable NMI/IRQ window open exits if needed */
+		if (vcpu->arch.nmi_pending)
+			kvm_x86_ops->enable_nmi_window(vcpu);
+		else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
+			kvm_x86_ops->enable_irq_window(vcpu);
+
+		if (kvm_lapic_enabled(vcpu)) {
+			update_cr8_intercept(vcpu);
+			kvm_lapic_sync_to_vapic(vcpu);
+		}
+	}
+
 	preempt_disable();
 
 	kvm_x86_ops->prepare_guest_switch(vcpu);
@@ -5053,35 +5067,17 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
 
 	local_irq_disable();
 
-	req_event = kvm_check_request(KVM_REQ_EVENT, vcpu);
-
 	if (!atomic_read(&vcpu->guest_mode) || vcpu->requests
 	    || need_resched() || signal_pending(current)) {
-		if (req_event)
-			kvm_make_request(KVM_REQ_EVENT, vcpu);
 		atomic_set(&vcpu->guest_mode, 0);
 		smp_wmb();
 		local_irq_enable();
 		preempt_enable();
+		kvm_x86_ops->cancel_injection(vcpu);
 		r = 1;
 		goto out;
 	}
 
-	if (req_event || req_int_win) {
-		inject_pending_event(vcpu);
-
-		/* enable NMI/IRQ window open exits if needed */
-		if (vcpu->arch.nmi_pending)
-			kvm_x86_ops->enable_nmi_window(vcpu);
-		else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
-			kvm_x86_ops->enable_irq_window(vcpu);
-
-		if (kvm_lapic_enabled(vcpu)) {
-			update_cr8_intercept(vcpu);
-			kvm_lapic_sync_to_vapic(vcpu);
-		}
-	}
-
 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
 
 	kvm_guest_enter();